RGLRE: Residual Graph Learning for circRNA-miRNA Interaction Prediction with Role-aware graph Embeddings

Abstract Circular RNAs (circRNAs) and microRNAs (miRNAs) are key regulators of gene expression, and their interactions (CMIs) are involved in many biological processes and diseases. Nevertheless, CMIs are still hard to validate experimentally because of limitations in resources, which is why solid computational models are required to obtain accurate CMIs prediction. This paper proposes a new Residual Graph Learning framework (RGLRE) for CMIs prediction using Role-aware Graph Embeddings. RGLRE combines sequence-based features extracted using DNABERT that encodes the nucleotide level patterns, and topological embeddings produced by Role2Vec which maintains the functional roles of the nodes in the interaction network. It employs a residual graph neural network that uses skip connections and batch normalization to suppress information loss and improve feature learning. RGLRE framework is evaluated on the benchmark datasets CMI-9905 and CMI-9589 using 5-fold cross-validation. It demonstrates superior performance compared to existing methods., with an accuracy of 0.9118 (CMI-9905) and 0.9269 (CMI-9589), alongside high AUROC (0.9650 and 0.9755) and AUPR (0.9791 and 0.9889) on the two benchmark datasets respectively. Comparative analysis shows that RGLRE outperforms current methods including boosting and graph-based deep learning methods. The framework's modular design and scalability make it a promising tool for broader biological interaction prediction tasks.